2 * linux/kernel/hrtimer.c
4 * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
5 * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
6 * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner
8 * High-resolution kernel timers
10 * In contrast to the low-resolution timeout API implemented in
11 * kernel/timer.c, hrtimers provide finer resolution and accuracy
12 * depending on system configuration and capabilities.
14 * These timers are currently used for:
18 * - precise in-kernel timing
20 * Started by: Thomas Gleixner and Ingo Molnar
23 * based on kernel/timer.c
25 * Help, testing, suggestions, bugfixes, improvements were
28 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
31 * For licencing details see kernel-base/COPYING
34 #include <linux/cpu.h>
35 #include <linux/export.h>
36 #include <linux/percpu.h>
37 #include <linux/hrtimer.h>
38 #include <linux/notifier.h>
39 #include <linux/syscalls.h>
40 #include <linux/kallsyms.h>
41 #include <linux/interrupt.h>
42 #include <linux/tick.h>
43 #include <linux/seq_file.h>
44 #include <linux/err.h>
45 #include <linux/debugobjects.h>
46 #include <linux/sched.h>
47 #include <linux/sched/sysctl.h>
48 #include <linux/sched/rt.h>
49 #include <linux/sched/deadline.h>
50 #include <linux/timer.h>
51 #include <linux/freezer.h>
53 #include <asm/uaccess.h>
55 #include <trace/events/timer.h>
57 #include "tick-internal.h"
62 * There are more clockids then hrtimer bases. Thus, we index
63 * into the timer bases by the hrtimer_base_type enum. When trying
64 * to reach a base using a clockid, hrtimer_clockid_to_base()
65 * is used to convert from clockid to the proper hrtimer_base_type.
67 DEFINE_PER_CPU(struct hrtimer_cpu_base
, hrtimer_bases
) =
69 .lock
= __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases
.lock
),
73 .index
= HRTIMER_BASE_MONOTONIC
,
74 .clockid
= CLOCK_MONOTONIC
,
75 .get_time
= &ktime_get
,
78 .index
= HRTIMER_BASE_REALTIME
,
79 .clockid
= CLOCK_REALTIME
,
80 .get_time
= &ktime_get_real
,
83 .index
= HRTIMER_BASE_BOOTTIME
,
84 .clockid
= CLOCK_BOOTTIME
,
85 .get_time
= &ktime_get_boottime
,
88 .index
= HRTIMER_BASE_TAI
,
90 .get_time
= &ktime_get_clocktai
,
95 static const int hrtimer_clock_to_base_table
[MAX_CLOCKS
] = {
96 [CLOCK_REALTIME
] = HRTIMER_BASE_REALTIME
,
97 [CLOCK_MONOTONIC
] = HRTIMER_BASE_MONOTONIC
,
98 [CLOCK_BOOTTIME
] = HRTIMER_BASE_BOOTTIME
,
99 [CLOCK_TAI
] = HRTIMER_BASE_TAI
,
102 static inline int hrtimer_clockid_to_base(clockid_t clock_id
)
104 return hrtimer_clock_to_base_table
[clock_id
];
109 * Get the coarse grained time at the softirq based on xtime and
112 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base
*base
)
114 ktime_t xtim
, mono
, boot
, tai
;
115 ktime_t off_real
, off_boot
, off_tai
;
117 mono
= ktime_get_update_offsets_tick(&off_real
, &off_boot
, &off_tai
);
118 boot
= ktime_add(mono
, off_boot
);
119 xtim
= ktime_add(mono
, off_real
);
120 tai
= ktime_add(mono
, off_tai
);
122 base
->clock_base
[HRTIMER_BASE_REALTIME
].softirq_time
= xtim
;
123 base
->clock_base
[HRTIMER_BASE_MONOTONIC
].softirq_time
= mono
;
124 base
->clock_base
[HRTIMER_BASE_BOOTTIME
].softirq_time
= boot
;
125 base
->clock_base
[HRTIMER_BASE_TAI
].softirq_time
= tai
;
129 * Functions and macros which are different for UP/SMP systems are kept in a
135 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
136 * means that all timers which are tied to this base via timer->base are
137 * locked, and the base itself is locked too.
139 * So __run_timers/migrate_timers can safely modify all timers which could
140 * be found on the lists/queues.
142 * When the timer's base is locked, and the timer removed from list, it is
143 * possible to set timer->base = NULL and drop the lock: the timer remains
147 struct hrtimer_clock_base
*lock_hrtimer_base(const struct hrtimer
*timer
,
148 unsigned long *flags
)
150 struct hrtimer_clock_base
*base
;
154 if (likely(base
!= NULL
)) {
155 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
156 if (likely(base
== timer
->base
))
158 /* The timer has migrated to another CPU: */
159 raw_spin_unlock_irqrestore(&base
->cpu_base
->lock
, *flags
);
166 * With HIGHRES=y we do not migrate the timer when it is expiring
167 * before the next event on the target cpu because we cannot reprogram
168 * the target cpu hardware and we would cause it to fire late.
170 * Called with cpu_base->lock of target cpu held.
173 hrtimer_check_target(struct hrtimer
*timer
, struct hrtimer_clock_base
*new_base
)
175 #ifdef CONFIG_HIGH_RES_TIMERS
178 if (!new_base
->cpu_base
->hres_active
)
181 expires
= ktime_sub(hrtimer_get_expires(timer
), new_base
->offset
);
182 return expires
.tv64
<= new_base
->cpu_base
->expires_next
.tv64
;
189 * Switch the timer base to the current CPU when possible.
191 static inline struct hrtimer_clock_base
*
192 switch_hrtimer_base(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
,
195 struct hrtimer_clock_base
*new_base
;
196 struct hrtimer_cpu_base
*new_cpu_base
;
197 int this_cpu
= smp_processor_id();
198 int cpu
= get_nohz_timer_target(pinned
);
199 int basenum
= base
->index
;
202 new_cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
203 new_base
= &new_cpu_base
->clock_base
[basenum
];
205 if (base
!= new_base
) {
207 * We are trying to move timer to new_base.
208 * However we can't change timer's base while it is running,
209 * so we keep it on the same CPU. No hassle vs. reprogramming
210 * the event source in the high resolution case. The softirq
211 * code will take care of this when the timer function has
212 * completed. There is no conflict as we hold the lock until
213 * the timer is enqueued.
215 if (unlikely(hrtimer_callback_running(timer
)))
218 /* See the comment in lock_timer_base() */
220 raw_spin_unlock(&base
->cpu_base
->lock
);
221 raw_spin_lock(&new_base
->cpu_base
->lock
);
223 if (cpu
!= this_cpu
&& hrtimer_check_target(timer
, new_base
)) {
225 raw_spin_unlock(&new_base
->cpu_base
->lock
);
226 raw_spin_lock(&base
->cpu_base
->lock
);
230 timer
->base
= new_base
;
232 if (cpu
!= this_cpu
&& hrtimer_check_target(timer
, new_base
)) {
240 #else /* CONFIG_SMP */
242 static inline struct hrtimer_clock_base
*
243 lock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
245 struct hrtimer_clock_base
*base
= timer
->base
;
247 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
252 # define switch_hrtimer_base(t, b, p) (b)
254 #endif /* !CONFIG_SMP */
257 * Functions for the union type storage format of ktime_t which are
258 * too large for inlining:
260 #if BITS_PER_LONG < 64
262 * Divide a ktime value by a nanosecond value
264 u64
__ktime_divns(const ktime_t kt
, s64 div
)
269 dclc
= ktime_to_ns(kt
);
270 /* Make sure the divisor is less than 2^32: */
276 do_div(dclc
, (unsigned long) div
);
280 EXPORT_SYMBOL_GPL(__ktime_divns
);
281 #endif /* BITS_PER_LONG >= 64 */
284 * Add two ktime values and do a safety check for overflow:
286 ktime_t
ktime_add_safe(const ktime_t lhs
, const ktime_t rhs
)
288 ktime_t res
= ktime_add(lhs
, rhs
);
291 * We use KTIME_SEC_MAX here, the maximum timeout which we can
292 * return to user space in a timespec:
294 if (res
.tv64
< 0 || res
.tv64
< lhs
.tv64
|| res
.tv64
< rhs
.tv64
)
295 res
= ktime_set(KTIME_SEC_MAX
, 0);
300 EXPORT_SYMBOL_GPL(ktime_add_safe
);
302 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
304 static struct debug_obj_descr hrtimer_debug_descr
;
306 static void *hrtimer_debug_hint(void *addr
)
308 return ((struct hrtimer
*) addr
)->function
;
312 * fixup_init is called when:
313 * - an active object is initialized
315 static int hrtimer_fixup_init(void *addr
, enum debug_obj_state state
)
317 struct hrtimer
*timer
= addr
;
320 case ODEBUG_STATE_ACTIVE
:
321 hrtimer_cancel(timer
);
322 debug_object_init(timer
, &hrtimer_debug_descr
);
330 * fixup_activate is called when:
331 * - an active object is activated
332 * - an unknown object is activated (might be a statically initialized object)
334 static int hrtimer_fixup_activate(void *addr
, enum debug_obj_state state
)
338 case ODEBUG_STATE_NOTAVAILABLE
:
342 case ODEBUG_STATE_ACTIVE
:
351 * fixup_free is called when:
352 * - an active object is freed
354 static int hrtimer_fixup_free(void *addr
, enum debug_obj_state state
)
356 struct hrtimer
*timer
= addr
;
359 case ODEBUG_STATE_ACTIVE
:
360 hrtimer_cancel(timer
);
361 debug_object_free(timer
, &hrtimer_debug_descr
);
368 static struct debug_obj_descr hrtimer_debug_descr
= {
370 .debug_hint
= hrtimer_debug_hint
,
371 .fixup_init
= hrtimer_fixup_init
,
372 .fixup_activate
= hrtimer_fixup_activate
,
373 .fixup_free
= hrtimer_fixup_free
,
376 static inline void debug_hrtimer_init(struct hrtimer
*timer
)
378 debug_object_init(timer
, &hrtimer_debug_descr
);
381 static inline void debug_hrtimer_activate(struct hrtimer
*timer
)
383 debug_object_activate(timer
, &hrtimer_debug_descr
);
386 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
)
388 debug_object_deactivate(timer
, &hrtimer_debug_descr
);
391 static inline void debug_hrtimer_free(struct hrtimer
*timer
)
393 debug_object_free(timer
, &hrtimer_debug_descr
);
396 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
397 enum hrtimer_mode mode
);
399 void hrtimer_init_on_stack(struct hrtimer
*timer
, clockid_t clock_id
,
400 enum hrtimer_mode mode
)
402 debug_object_init_on_stack(timer
, &hrtimer_debug_descr
);
403 __hrtimer_init(timer
, clock_id
, mode
);
405 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack
);
407 void destroy_hrtimer_on_stack(struct hrtimer
*timer
)
409 debug_object_free(timer
, &hrtimer_debug_descr
);
413 static inline void debug_hrtimer_init(struct hrtimer
*timer
) { }
414 static inline void debug_hrtimer_activate(struct hrtimer
*timer
) { }
415 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
) { }
419 debug_init(struct hrtimer
*timer
, clockid_t clockid
,
420 enum hrtimer_mode mode
)
422 debug_hrtimer_init(timer
);
423 trace_hrtimer_init(timer
, clockid
, mode
);
426 static inline void debug_activate(struct hrtimer
*timer
)
428 debug_hrtimer_activate(timer
);
429 trace_hrtimer_start(timer
);
432 static inline void debug_deactivate(struct hrtimer
*timer
)
434 debug_hrtimer_deactivate(timer
);
435 trace_hrtimer_cancel(timer
);
438 #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS)
439 static ktime_t
__hrtimer_get_next_event(struct hrtimer_cpu_base
*cpu_base
)
441 struct hrtimer_clock_base
*base
= cpu_base
->clock_base
;
442 ktime_t expires
, expires_next
= { .tv64
= KTIME_MAX
};
445 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++, base
++) {
446 struct timerqueue_node
*next
;
447 struct hrtimer
*timer
;
449 next
= timerqueue_getnext(&base
->active
);
453 timer
= container_of(next
, struct hrtimer
, node
);
454 expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
455 if (expires
.tv64
< expires_next
.tv64
)
456 expires_next
= expires
;
459 * clock_was_set() might have changed base->offset of any of
460 * the clock bases so the result might be negative. Fix it up
461 * to prevent a false positive in clockevents_program_event().
463 if (expires_next
.tv64
< 0)
464 expires_next
.tv64
= 0;
469 /* High resolution timer related functions */
470 #ifdef CONFIG_HIGH_RES_TIMERS
473 * High resolution timer enabled ?
475 static int hrtimer_hres_enabled __read_mostly
= 1;
476 unsigned int hrtimer_resolution __read_mostly
= LOW_RES_NSEC
;
477 EXPORT_SYMBOL_GPL(hrtimer_resolution
);
480 * Enable / Disable high resolution mode
482 static int __init
setup_hrtimer_hres(char *str
)
484 if (!strcmp(str
, "off"))
485 hrtimer_hres_enabled
= 0;
486 else if (!strcmp(str
, "on"))
487 hrtimer_hres_enabled
= 1;
493 __setup("highres=", setup_hrtimer_hres
);
496 * hrtimer_high_res_enabled - query, if the highres mode is enabled
498 static inline int hrtimer_is_hres_enabled(void)
500 return hrtimer_hres_enabled
;
504 * Is the high resolution mode active ?
506 static inline int hrtimer_hres_active(void)
508 return __this_cpu_read(hrtimer_bases
.hres_active
);
512 * Reprogram the event source with checking both queues for the
514 * Called with interrupts disabled and base->lock held
517 hrtimer_force_reprogram(struct hrtimer_cpu_base
*cpu_base
, int skip_equal
)
519 ktime_t expires_next
= __hrtimer_get_next_event(cpu_base
);
521 if (skip_equal
&& expires_next
.tv64
== cpu_base
->expires_next
.tv64
)
524 cpu_base
->expires_next
.tv64
= expires_next
.tv64
;
527 * If a hang was detected in the last timer interrupt then we
528 * leave the hang delay active in the hardware. We want the
529 * system to make progress. That also prevents the following
531 * T1 expires 50ms from now
532 * T2 expires 5s from now
534 * T1 is removed, so this code is called and would reprogram
535 * the hardware to 5s from now. Any hrtimer_start after that
536 * will not reprogram the hardware due to hang_detected being
537 * set. So we'd effectivly block all timers until the T2 event
540 if (cpu_base
->hang_detected
)
543 if (cpu_base
->expires_next
.tv64
!= KTIME_MAX
)
544 tick_program_event(cpu_base
->expires_next
, 1);
548 * Shared reprogramming for clock_realtime and clock_monotonic
550 * When a timer is enqueued and expires earlier than the already enqueued
551 * timers, we have to check, whether it expires earlier than the timer for
552 * which the clock event device was armed.
554 * Note, that in case the state has HRTIMER_STATE_CALLBACK set, no reprogramming
555 * and no expiry check happens. The timer gets enqueued into the rbtree. The
556 * reprogramming and expiry check is done in the hrtimer_interrupt or in the
559 * Called with interrupts disabled and base->cpu_base.lock held
561 static int hrtimer_reprogram(struct hrtimer
*timer
,
562 struct hrtimer_clock_base
*base
)
564 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
565 ktime_t expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
568 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer
) < 0);
571 * When the callback is running, we do not reprogram the clock event
572 * device. The timer callback is either running on a different CPU or
573 * the callback is executed in the hrtimer_interrupt context. The
574 * reprogramming is handled either by the softirq, which called the
575 * callback or at the end of the hrtimer_interrupt.
577 if (hrtimer_callback_running(timer
))
581 * CLOCK_REALTIME timer might be requested with an absolute
582 * expiry time which is less than base->offset. Nothing wrong
583 * about that, just avoid to call into the tick code, which
584 * has now objections against negative expiry values.
586 if (expires
.tv64
< 0)
589 if (expires
.tv64
>= cpu_base
->expires_next
.tv64
)
593 * When the target cpu of the timer is currently executing
594 * hrtimer_interrupt(), then we do not touch the clock event
595 * device. hrtimer_interrupt() will reevaluate all clock bases
596 * before reprogramming the device.
598 if (cpu_base
->in_hrtirq
)
602 * If a hang was detected in the last timer interrupt then we
603 * do not schedule a timer which is earlier than the expiry
604 * which we enforced in the hang detection. We want the system
607 if (cpu_base
->hang_detected
)
611 * Clockevents returns -ETIME, when the event was in the past.
613 res
= tick_program_event(expires
, 0);
614 if (!IS_ERR_VALUE(res
))
615 cpu_base
->expires_next
= expires
;
620 * Initialize the high resolution related parts of cpu_base
622 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
)
624 base
->expires_next
.tv64
= KTIME_MAX
;
625 base
->hres_active
= 0;
628 static inline ktime_t
hrtimer_update_base(struct hrtimer_cpu_base
*base
)
630 ktime_t
*offs_real
= &base
->clock_base
[HRTIMER_BASE_REALTIME
].offset
;
631 ktime_t
*offs_boot
= &base
->clock_base
[HRTIMER_BASE_BOOTTIME
].offset
;
632 ktime_t
*offs_tai
= &base
->clock_base
[HRTIMER_BASE_TAI
].offset
;
634 return ktime_get_update_offsets_now(offs_real
, offs_boot
, offs_tai
);
638 * Retrigger next event is called after clock was set
640 * Called with interrupts disabled via on_each_cpu()
642 static void retrigger_next_event(void *arg
)
644 struct hrtimer_cpu_base
*base
= this_cpu_ptr(&hrtimer_bases
);
646 if (!hrtimer_hres_active())
649 raw_spin_lock(&base
->lock
);
650 hrtimer_update_base(base
);
651 hrtimer_force_reprogram(base
, 0);
652 raw_spin_unlock(&base
->lock
);
656 * Switch to high resolution mode
658 static int hrtimer_switch_to_hres(void)
660 int cpu
= smp_processor_id();
661 struct hrtimer_cpu_base
*base
= &per_cpu(hrtimer_bases
, cpu
);
664 if (base
->hres_active
)
667 local_irq_save(flags
);
669 if (tick_init_highres()) {
670 local_irq_restore(flags
);
671 printk(KERN_WARNING
"Could not switch to high resolution "
672 "mode on CPU %d\n", cpu
);
675 base
->hres_active
= 1;
676 hrtimer_resolution
= HIGH_RES_NSEC
;
678 tick_setup_sched_timer();
679 /* "Retrigger" the interrupt to get things going */
680 retrigger_next_event(NULL
);
681 local_irq_restore(flags
);
685 static void clock_was_set_work(struct work_struct
*work
)
690 static DECLARE_WORK(hrtimer_work
, clock_was_set_work
);
693 * Called from timekeeping and resume code to reprogramm the hrtimer
694 * interrupt device on all cpus.
696 void clock_was_set_delayed(void)
698 schedule_work(&hrtimer_work
);
703 static inline int hrtimer_hres_active(void) { return 0; }
704 static inline int hrtimer_is_hres_enabled(void) { return 0; }
705 static inline int hrtimer_switch_to_hres(void) { return 0; }
707 hrtimer_force_reprogram(struct hrtimer_cpu_base
*base
, int skip_equal
) { }
708 static inline int hrtimer_reprogram(struct hrtimer
*timer
,
709 struct hrtimer_clock_base
*base
)
713 static inline void hrtimer_init_hres(struct hrtimer_cpu_base
*base
) { }
714 static inline void retrigger_next_event(void *arg
) { }
716 #endif /* CONFIG_HIGH_RES_TIMERS */
719 * Clock realtime was set
721 * Change the offset of the realtime clock vs. the monotonic
724 * We might have to reprogram the high resolution timer interrupt. On
725 * SMP we call the architecture specific code to retrigger _all_ high
726 * resolution timer interrupts. On UP we just disable interrupts and
727 * call the high resolution interrupt code.
729 void clock_was_set(void)
731 #ifdef CONFIG_HIGH_RES_TIMERS
732 /* Retrigger the CPU local events everywhere */
733 on_each_cpu(retrigger_next_event
, NULL
, 1);
735 timerfd_clock_was_set();
739 * During resume we might have to reprogram the high resolution timer
740 * interrupt on all online CPUs. However, all other CPUs will be
741 * stopped with IRQs interrupts disabled so the clock_was_set() call
744 void hrtimers_resume(void)
746 WARN_ONCE(!irqs_disabled(),
747 KERN_INFO
"hrtimers_resume() called with IRQs enabled!");
749 /* Retrigger on the local CPU */
750 retrigger_next_event(NULL
);
751 /* And schedule a retrigger for all others */
752 clock_was_set_delayed();
755 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer
*timer
)
757 #ifdef CONFIG_TIMER_STATS
758 if (timer
->start_site
)
760 timer
->start_site
= __builtin_return_address(0);
761 memcpy(timer
->start_comm
, current
->comm
, TASK_COMM_LEN
);
762 timer
->start_pid
= current
->pid
;
766 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer
*timer
)
768 #ifdef CONFIG_TIMER_STATS
769 timer
->start_site
= NULL
;
773 static inline void timer_stats_account_hrtimer(struct hrtimer
*timer
)
775 #ifdef CONFIG_TIMER_STATS
776 if (likely(!timer_stats_active
))
778 timer_stats_update_stats(timer
, timer
->start_pid
, timer
->start_site
,
779 timer
->function
, timer
->start_comm
, 0);
784 * Counterpart to lock_hrtimer_base above:
787 void unlock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
789 raw_spin_unlock_irqrestore(&timer
->base
->cpu_base
->lock
, *flags
);
793 * hrtimer_forward - forward the timer expiry
794 * @timer: hrtimer to forward
795 * @now: forward past this time
796 * @interval: the interval to forward
798 * Forward the timer expiry so it will expire in the future.
799 * Returns the number of overruns.
801 * Can be safely called from the callback function of @timer. If
802 * called from other contexts @timer must neither be enqueued nor
803 * running the callback and the caller needs to take care of
806 * Note: This only updates the timer expiry value and does not requeue
809 u64
hrtimer_forward(struct hrtimer
*timer
, ktime_t now
, ktime_t interval
)
814 delta
= ktime_sub(now
, hrtimer_get_expires(timer
));
819 if (interval
.tv64
< hrtimer_resolution
)
820 interval
.tv64
= hrtimer_resolution
;
822 if (unlikely(delta
.tv64
>= interval
.tv64
)) {
823 s64 incr
= ktime_to_ns(interval
);
825 orun
= ktime_divns(delta
, incr
);
826 hrtimer_add_expires_ns(timer
, incr
* orun
);
827 if (hrtimer_get_expires_tv64(timer
) > now
.tv64
)
830 * This (and the ktime_add() below) is the
831 * correction for exact:
835 hrtimer_add_expires(timer
, interval
);
839 EXPORT_SYMBOL_GPL(hrtimer_forward
);
842 * enqueue_hrtimer - internal function to (re)start a timer
844 * The timer is inserted in expiry order. Insertion into the
845 * red black tree is O(log(n)). Must hold the base lock.
847 * Returns 1 when the new timer is the leftmost timer in the tree.
849 static int enqueue_hrtimer(struct hrtimer
*timer
,
850 struct hrtimer_clock_base
*base
)
852 debug_activate(timer
);
854 timerqueue_add(&base
->active
, &timer
->node
);
855 base
->cpu_base
->active_bases
|= 1 << base
->index
;
858 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
859 * state of a possibly running callback.
861 timer
->state
|= HRTIMER_STATE_ENQUEUED
;
863 return (&timer
->node
== base
->active
.next
);
867 * __remove_hrtimer - internal function to remove a timer
869 * Caller must hold the base lock.
871 * High resolution timer mode reprograms the clock event device when the
872 * timer is the one which expires next. The caller can disable this by setting
873 * reprogram to zero. This is useful, when the context does a reprogramming
874 * anyway (e.g. timer interrupt)
876 static void __remove_hrtimer(struct hrtimer
*timer
,
877 struct hrtimer_clock_base
*base
,
878 unsigned long newstate
, int reprogram
)
880 struct timerqueue_node
*next_timer
;
881 if (!(timer
->state
& HRTIMER_STATE_ENQUEUED
))
884 next_timer
= timerqueue_getnext(&base
->active
);
885 timerqueue_del(&base
->active
, &timer
->node
);
886 if (!timerqueue_getnext(&base
->active
))
887 base
->cpu_base
->active_bases
&= ~(1 << base
->index
);
889 if (&timer
->node
== next_timer
) {
890 #ifdef CONFIG_HIGH_RES_TIMERS
891 /* Reprogram the clock event device. if enabled */
892 if (reprogram
&& hrtimer_hres_active()) {
895 expires
= ktime_sub(hrtimer_get_expires(timer
),
897 if (base
->cpu_base
->expires_next
.tv64
== expires
.tv64
)
898 hrtimer_force_reprogram(base
->cpu_base
, 1);
903 timer
->state
= newstate
;
907 * remove hrtimer, called with base lock held
910 remove_hrtimer(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
)
912 if (hrtimer_is_queued(timer
)) {
917 * Remove the timer and force reprogramming when high
918 * resolution mode is active and the timer is on the current
919 * CPU. If we remove a timer on another CPU, reprogramming is
920 * skipped. The interrupt event on this CPU is fired and
921 * reprogramming happens in the interrupt handler. This is a
922 * rare case and less expensive than a smp call.
924 debug_deactivate(timer
);
925 timer_stats_hrtimer_clear_start_info(timer
);
926 reprogram
= base
->cpu_base
== this_cpu_ptr(&hrtimer_bases
);
928 * We must preserve the CALLBACK state flag here,
929 * otherwise we could move the timer base in
930 * switch_hrtimer_base.
932 state
= timer
->state
& HRTIMER_STATE_CALLBACK
;
933 __remove_hrtimer(timer
, base
, state
, reprogram
);
939 int __hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
940 unsigned long delta_ns
, const enum hrtimer_mode mode
,
943 struct hrtimer_clock_base
*base
, *new_base
;
947 base
= lock_hrtimer_base(timer
, &flags
);
949 /* Remove an active timer from the queue: */
950 ret
= remove_hrtimer(timer
, base
);
952 if (mode
& HRTIMER_MODE_REL
) {
953 tim
= ktime_add_safe(tim
, base
->get_time());
955 * CONFIG_TIME_LOW_RES is a temporary way for architectures
956 * to signal that they simply return xtime in
957 * do_gettimeoffset(). In this case we want to round up by
958 * resolution when starting a relative timer, to avoid short
959 * timeouts. This will go away with the GTOD framework.
961 #ifdef CONFIG_TIME_LOW_RES
962 tim
= ktime_add_safe(tim
, ktime_set(0, hrtimer_resolution
));
966 hrtimer_set_expires_range_ns(timer
, tim
, delta_ns
);
968 /* Switch the timer base, if necessary: */
969 new_base
= switch_hrtimer_base(timer
, base
, mode
& HRTIMER_MODE_PINNED
);
971 timer_stats_hrtimer_set_start_info(timer
);
973 leftmost
= enqueue_hrtimer(timer
, new_base
);
976 unlock_hrtimer_base(timer
, &flags
);
980 if (!hrtimer_is_hres_active(timer
)) {
982 * Kick to reschedule the next tick to handle the new timer
983 * on dynticks target.
985 wake_up_nohz_cpu(new_base
->cpu_base
->cpu
);
986 } else if (new_base
->cpu_base
== this_cpu_ptr(&hrtimer_bases
) &&
987 hrtimer_reprogram(timer
, new_base
)) {
989 * Only allow reprogramming if the new base is on this CPU.
990 * (it might still be on another CPU if the timer was pending)
992 * XXX send_remote_softirq() ?
996 * We need to drop cpu_base->lock to avoid a
997 * lock ordering issue vs. rq->lock.
999 raw_spin_unlock(&new_base
->cpu_base
->lock
);
1000 raise_softirq_irqoff(HRTIMER_SOFTIRQ
);
1001 local_irq_restore(flags
);
1004 __raise_softirq_irqoff(HRTIMER_SOFTIRQ
);
1008 unlock_hrtimer_base(timer
, &flags
);
1012 EXPORT_SYMBOL_GPL(__hrtimer_start_range_ns
);
1015 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1016 * @timer: the timer to be added
1018 * @delta_ns: "slack" range for the timer
1019 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1020 * relative (HRTIMER_MODE_REL)
1024 * 1 when the timer was active
1026 int hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
1027 unsigned long delta_ns
, const enum hrtimer_mode mode
)
1029 return __hrtimer_start_range_ns(timer
, tim
, delta_ns
, mode
, 1);
1031 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns
);
1034 * hrtimer_start - (re)start an hrtimer on the current CPU
1035 * @timer: the timer to be added
1037 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1038 * relative (HRTIMER_MODE_REL)
1042 * 1 when the timer was active
1045 hrtimer_start(struct hrtimer
*timer
, ktime_t tim
, const enum hrtimer_mode mode
)
1047 return __hrtimer_start_range_ns(timer
, tim
, 0, mode
, 1);
1049 EXPORT_SYMBOL_GPL(hrtimer_start
);
1053 * hrtimer_try_to_cancel - try to deactivate a timer
1054 * @timer: hrtimer to stop
1057 * 0 when the timer was not active
1058 * 1 when the timer was active
1059 * -1 when the timer is currently excuting the callback function and
1062 int hrtimer_try_to_cancel(struct hrtimer
*timer
)
1064 struct hrtimer_clock_base
*base
;
1065 unsigned long flags
;
1068 base
= lock_hrtimer_base(timer
, &flags
);
1070 if (!hrtimer_callback_running(timer
))
1071 ret
= remove_hrtimer(timer
, base
);
1073 unlock_hrtimer_base(timer
, &flags
);
1078 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel
);
1081 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1082 * @timer: the timer to be cancelled
1085 * 0 when the timer was not active
1086 * 1 when the timer was active
1088 int hrtimer_cancel(struct hrtimer
*timer
)
1091 int ret
= hrtimer_try_to_cancel(timer
);
1098 EXPORT_SYMBOL_GPL(hrtimer_cancel
);
1101 * hrtimer_get_remaining - get remaining time for the timer
1102 * @timer: the timer to read
1104 ktime_t
hrtimer_get_remaining(const struct hrtimer
*timer
)
1106 unsigned long flags
;
1109 lock_hrtimer_base(timer
, &flags
);
1110 rem
= hrtimer_expires_remaining(timer
);
1111 unlock_hrtimer_base(timer
, &flags
);
1115 EXPORT_SYMBOL_GPL(hrtimer_get_remaining
);
1117 #ifdef CONFIG_NO_HZ_COMMON
1119 * hrtimer_get_next_event - get the time until next expiry event
1121 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1124 ktime_t
hrtimer_get_next_event(void)
1126 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1127 ktime_t mindelta
= { .tv64
= KTIME_MAX
};
1128 unsigned long flags
;
1130 raw_spin_lock_irqsave(&cpu_base
->lock
, flags
);
1132 if (!hrtimer_hres_active())
1133 mindelta
= ktime_sub(__hrtimer_get_next_event(cpu_base
),
1136 raw_spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
1138 if (mindelta
.tv64
< 0)
1144 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1145 enum hrtimer_mode mode
)
1147 struct hrtimer_cpu_base
*cpu_base
;
1150 memset(timer
, 0, sizeof(struct hrtimer
));
1152 cpu_base
= raw_cpu_ptr(&hrtimer_bases
);
1154 if (clock_id
== CLOCK_REALTIME
&& mode
!= HRTIMER_MODE_ABS
)
1155 clock_id
= CLOCK_MONOTONIC
;
1157 base
= hrtimer_clockid_to_base(clock_id
);
1158 timer
->base
= &cpu_base
->clock_base
[base
];
1159 timerqueue_init(&timer
->node
);
1161 #ifdef CONFIG_TIMER_STATS
1162 timer
->start_site
= NULL
;
1163 timer
->start_pid
= -1;
1164 memset(timer
->start_comm
, 0, TASK_COMM_LEN
);
1169 * hrtimer_init - initialize a timer to the given clock
1170 * @timer: the timer to be initialized
1171 * @clock_id: the clock to be used
1172 * @mode: timer mode abs/rel
1174 void hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1175 enum hrtimer_mode mode
)
1177 debug_init(timer
, clock_id
, mode
);
1178 __hrtimer_init(timer
, clock_id
, mode
);
1180 EXPORT_SYMBOL_GPL(hrtimer_init
);
1182 static void __run_hrtimer(struct hrtimer
*timer
, ktime_t
*now
)
1184 struct hrtimer_clock_base
*base
= timer
->base
;
1185 struct hrtimer_cpu_base
*cpu_base
= base
->cpu_base
;
1186 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1189 WARN_ON(!irqs_disabled());
1191 debug_deactivate(timer
);
1192 __remove_hrtimer(timer
, base
, HRTIMER_STATE_CALLBACK
, 0);
1193 timer_stats_account_hrtimer(timer
);
1194 fn
= timer
->function
;
1197 * Because we run timers from hardirq context, there is no chance
1198 * they get migrated to another cpu, therefore its safe to unlock
1201 raw_spin_unlock(&cpu_base
->lock
);
1202 trace_hrtimer_expire_entry(timer
, now
);
1203 restart
= fn(timer
);
1204 trace_hrtimer_expire_exit(timer
);
1205 raw_spin_lock(&cpu_base
->lock
);
1208 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1209 * we do not reprogramm the event hardware. Happens either in
1210 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1212 if (restart
!= HRTIMER_NORESTART
) {
1213 BUG_ON(timer
->state
!= HRTIMER_STATE_CALLBACK
);
1214 enqueue_hrtimer(timer
, base
);
1217 WARN_ON_ONCE(!(timer
->state
& HRTIMER_STATE_CALLBACK
));
1219 timer
->state
&= ~HRTIMER_STATE_CALLBACK
;
1222 #ifdef CONFIG_HIGH_RES_TIMERS
1225 * High resolution timer interrupt
1226 * Called with interrupts disabled
1228 void hrtimer_interrupt(struct clock_event_device
*dev
)
1230 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1231 ktime_t expires_next
, now
, entry_time
, delta
;
1234 BUG_ON(!cpu_base
->hres_active
);
1235 cpu_base
->nr_events
++;
1236 dev
->next_event
.tv64
= KTIME_MAX
;
1238 raw_spin_lock(&cpu_base
->lock
);
1239 entry_time
= now
= hrtimer_update_base(cpu_base
);
1241 cpu_base
->in_hrtirq
= 1;
1243 * We set expires_next to KTIME_MAX here with cpu_base->lock
1244 * held to prevent that a timer is enqueued in our queue via
1245 * the migration code. This does not affect enqueueing of
1246 * timers which run their callback and need to be requeued on
1249 cpu_base
->expires_next
.tv64
= KTIME_MAX
;
1251 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1252 struct hrtimer_clock_base
*base
;
1253 struct timerqueue_node
*node
;
1256 if (!(cpu_base
->active_bases
& (1 << i
)))
1259 base
= cpu_base
->clock_base
+ i
;
1260 basenow
= ktime_add(now
, base
->offset
);
1262 while ((node
= timerqueue_getnext(&base
->active
))) {
1263 struct hrtimer
*timer
;
1265 timer
= container_of(node
, struct hrtimer
, node
);
1268 * The immediate goal for using the softexpires is
1269 * minimizing wakeups, not running timers at the
1270 * earliest interrupt after their soft expiration.
1271 * This allows us to avoid using a Priority Search
1272 * Tree, which can answer a stabbing querry for
1273 * overlapping intervals and instead use the simple
1274 * BST we already have.
1275 * We don't add extra wakeups by delaying timers that
1276 * are right-of a not yet expired timer, because that
1277 * timer will have to trigger a wakeup anyway.
1279 if (basenow
.tv64
< hrtimer_get_softexpires_tv64(timer
))
1282 __run_hrtimer(timer
, &basenow
);
1285 /* Reevaluate the clock bases for the next expiry */
1286 expires_next
= __hrtimer_get_next_event(cpu_base
);
1288 * Store the new expiry value so the migration code can verify
1291 cpu_base
->expires_next
= expires_next
;
1292 cpu_base
->in_hrtirq
= 0;
1293 raw_spin_unlock(&cpu_base
->lock
);
1295 /* Reprogramming necessary ? */
1296 if (expires_next
.tv64
== KTIME_MAX
||
1297 !tick_program_event(expires_next
, 0)) {
1298 cpu_base
->hang_detected
= 0;
1303 * The next timer was already expired due to:
1305 * - long lasting callbacks
1306 * - being scheduled away when running in a VM
1308 * We need to prevent that we loop forever in the hrtimer
1309 * interrupt routine. We give it 3 attempts to avoid
1310 * overreacting on some spurious event.
1312 * Acquire base lock for updating the offsets and retrieving
1315 raw_spin_lock(&cpu_base
->lock
);
1316 now
= hrtimer_update_base(cpu_base
);
1317 cpu_base
->nr_retries
++;
1321 * Give the system a chance to do something else than looping
1322 * here. We stored the entry time, so we know exactly how long
1323 * we spent here. We schedule the next event this amount of
1326 cpu_base
->nr_hangs
++;
1327 cpu_base
->hang_detected
= 1;
1328 raw_spin_unlock(&cpu_base
->lock
);
1329 delta
= ktime_sub(now
, entry_time
);
1330 if ((unsigned int)delta
.tv64
> cpu_base
->max_hang_time
)
1331 cpu_base
->max_hang_time
= (unsigned int) delta
.tv64
;
1333 * Limit it to a sensible value as we enforce a longer
1334 * delay. Give the CPU at least 100ms to catch up.
1336 if (delta
.tv64
> 100 * NSEC_PER_MSEC
)
1337 expires_next
= ktime_add_ns(now
, 100 * NSEC_PER_MSEC
);
1339 expires_next
= ktime_add(now
, delta
);
1340 tick_program_event(expires_next
, 1);
1341 printk_once(KERN_WARNING
"hrtimer: interrupt took %llu ns\n",
1342 ktime_to_ns(delta
));
1346 * local version of hrtimer_peek_ahead_timers() called with interrupts
1349 static void __hrtimer_peek_ahead_timers(void)
1351 struct tick_device
*td
;
1353 if (!hrtimer_hres_active())
1356 td
= this_cpu_ptr(&tick_cpu_device
);
1357 if (td
&& td
->evtdev
)
1358 hrtimer_interrupt(td
->evtdev
);
1362 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1364 * hrtimer_peek_ahead_timers will peek at the timer queue of
1365 * the current cpu and check if there are any timers for which
1366 * the soft expires time has passed. If any such timers exist,
1367 * they are run immediately and then removed from the timer queue.
1370 void hrtimer_peek_ahead_timers(void)
1372 unsigned long flags
;
1374 local_irq_save(flags
);
1375 __hrtimer_peek_ahead_timers();
1376 local_irq_restore(flags
);
1379 static void run_hrtimer_softirq(struct softirq_action
*h
)
1381 hrtimer_peek_ahead_timers();
1384 #else /* CONFIG_HIGH_RES_TIMERS */
1386 static inline void __hrtimer_peek_ahead_timers(void) { }
1388 #endif /* !CONFIG_HIGH_RES_TIMERS */
1391 * Called from timer softirq every jiffy, expire hrtimers:
1393 * For HRT its the fall back code to run the softirq in the timer
1394 * softirq context in case the hrtimer initialization failed or has
1395 * not been done yet.
1397 void hrtimer_run_pending(void)
1399 if (hrtimer_hres_active())
1403 * This _is_ ugly: We have to check in the softirq context,
1404 * whether we can switch to highres and / or nohz mode. The
1405 * clocksource switch happens in the timer interrupt with
1406 * xtime_lock held. Notification from there only sets the
1407 * check bit in the tick_oneshot code, otherwise we might
1408 * deadlock vs. xtime_lock.
1410 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1411 hrtimer_switch_to_hres();
1415 * Called from hardirq context every jiffy
1417 void hrtimer_run_queues(void)
1419 struct timerqueue_node
*node
;
1420 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1421 struct hrtimer_clock_base
*base
;
1422 int index
, gettime
= 1;
1424 if (hrtimer_hres_active())
1427 for (index
= 0; index
< HRTIMER_MAX_CLOCK_BASES
; index
++) {
1428 base
= &cpu_base
->clock_base
[index
];
1429 if (!timerqueue_getnext(&base
->active
))
1433 hrtimer_get_softirq_time(cpu_base
);
1437 raw_spin_lock(&cpu_base
->lock
);
1439 while ((node
= timerqueue_getnext(&base
->active
))) {
1440 struct hrtimer
*timer
;
1442 timer
= container_of(node
, struct hrtimer
, node
);
1443 if (base
->softirq_time
.tv64
<=
1444 hrtimer_get_expires_tv64(timer
))
1447 __run_hrtimer(timer
, &base
->softirq_time
);
1449 raw_spin_unlock(&cpu_base
->lock
);
1454 * Sleep related functions:
1456 static enum hrtimer_restart
hrtimer_wakeup(struct hrtimer
*timer
)
1458 struct hrtimer_sleeper
*t
=
1459 container_of(timer
, struct hrtimer_sleeper
, timer
);
1460 struct task_struct
*task
= t
->task
;
1464 wake_up_process(task
);
1466 return HRTIMER_NORESTART
;
1469 void hrtimer_init_sleeper(struct hrtimer_sleeper
*sl
, struct task_struct
*task
)
1471 sl
->timer
.function
= hrtimer_wakeup
;
1474 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper
);
1476 static int __sched
do_nanosleep(struct hrtimer_sleeper
*t
, enum hrtimer_mode mode
)
1478 hrtimer_init_sleeper(t
, current
);
1481 set_current_state(TASK_INTERRUPTIBLE
);
1482 hrtimer_start_expires(&t
->timer
, mode
);
1483 if (!hrtimer_active(&t
->timer
))
1486 if (likely(t
->task
))
1487 freezable_schedule();
1489 hrtimer_cancel(&t
->timer
);
1490 mode
= HRTIMER_MODE_ABS
;
1492 } while (t
->task
&& !signal_pending(current
));
1494 __set_current_state(TASK_RUNNING
);
1496 return t
->task
== NULL
;
1499 static int update_rmtp(struct hrtimer
*timer
, struct timespec __user
*rmtp
)
1501 struct timespec rmt
;
1504 rem
= hrtimer_expires_remaining(timer
);
1507 rmt
= ktime_to_timespec(rem
);
1509 if (copy_to_user(rmtp
, &rmt
, sizeof(*rmtp
)))
1515 long __sched
hrtimer_nanosleep_restart(struct restart_block
*restart
)
1517 struct hrtimer_sleeper t
;
1518 struct timespec __user
*rmtp
;
1521 hrtimer_init_on_stack(&t
.timer
, restart
->nanosleep
.clockid
,
1523 hrtimer_set_expires_tv64(&t
.timer
, restart
->nanosleep
.expires
);
1525 if (do_nanosleep(&t
, HRTIMER_MODE_ABS
))
1528 rmtp
= restart
->nanosleep
.rmtp
;
1530 ret
= update_rmtp(&t
.timer
, rmtp
);
1535 /* The other values in restart are already filled in */
1536 ret
= -ERESTART_RESTARTBLOCK
;
1538 destroy_hrtimer_on_stack(&t
.timer
);
1542 long hrtimer_nanosleep(struct timespec
*rqtp
, struct timespec __user
*rmtp
,
1543 const enum hrtimer_mode mode
, const clockid_t clockid
)
1545 struct restart_block
*restart
;
1546 struct hrtimer_sleeper t
;
1548 unsigned long slack
;
1550 slack
= current
->timer_slack_ns
;
1551 if (dl_task(current
) || rt_task(current
))
1554 hrtimer_init_on_stack(&t
.timer
, clockid
, mode
);
1555 hrtimer_set_expires_range_ns(&t
.timer
, timespec_to_ktime(*rqtp
), slack
);
1556 if (do_nanosleep(&t
, mode
))
1559 /* Absolute timers do not update the rmtp value and restart: */
1560 if (mode
== HRTIMER_MODE_ABS
) {
1561 ret
= -ERESTARTNOHAND
;
1566 ret
= update_rmtp(&t
.timer
, rmtp
);
1571 restart
= ¤t
->restart_block
;
1572 restart
->fn
= hrtimer_nanosleep_restart
;
1573 restart
->nanosleep
.clockid
= t
.timer
.base
->clockid
;
1574 restart
->nanosleep
.rmtp
= rmtp
;
1575 restart
->nanosleep
.expires
= hrtimer_get_expires_tv64(&t
.timer
);
1577 ret
= -ERESTART_RESTARTBLOCK
;
1579 destroy_hrtimer_on_stack(&t
.timer
);
1583 SYSCALL_DEFINE2(nanosleep
, struct timespec __user
*, rqtp
,
1584 struct timespec __user
*, rmtp
)
1588 if (copy_from_user(&tu
, rqtp
, sizeof(tu
)))
1591 if (!timespec_valid(&tu
))
1594 return hrtimer_nanosleep(&tu
, rmtp
, HRTIMER_MODE_REL
, CLOCK_MONOTONIC
);
1598 * Functions related to boot-time initialization:
1600 static void init_hrtimers_cpu(int cpu
)
1602 struct hrtimer_cpu_base
*cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
1605 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1606 cpu_base
->clock_base
[i
].cpu_base
= cpu_base
;
1607 timerqueue_init_head(&cpu_base
->clock_base
[i
].active
);
1610 cpu_base
->cpu
= cpu
;
1611 hrtimer_init_hres(cpu_base
);
1614 #ifdef CONFIG_HOTPLUG_CPU
1616 static void migrate_hrtimer_list(struct hrtimer_clock_base
*old_base
,
1617 struct hrtimer_clock_base
*new_base
)
1619 struct hrtimer
*timer
;
1620 struct timerqueue_node
*node
;
1622 while ((node
= timerqueue_getnext(&old_base
->active
))) {
1623 timer
= container_of(node
, struct hrtimer
, node
);
1624 BUG_ON(hrtimer_callback_running(timer
));
1625 debug_deactivate(timer
);
1628 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1629 * timer could be seen as !active and just vanish away
1630 * under us on another CPU
1632 __remove_hrtimer(timer
, old_base
, HRTIMER_STATE_MIGRATE
, 0);
1633 timer
->base
= new_base
;
1635 * Enqueue the timers on the new cpu. This does not
1636 * reprogram the event device in case the timer
1637 * expires before the earliest on this CPU, but we run
1638 * hrtimer_interrupt after we migrated everything to
1639 * sort out already expired timers and reprogram the
1642 enqueue_hrtimer(timer
, new_base
);
1644 /* Clear the migration state bit */
1645 timer
->state
&= ~HRTIMER_STATE_MIGRATE
;
1649 static void migrate_hrtimers(int scpu
)
1651 struct hrtimer_cpu_base
*old_base
, *new_base
;
1654 BUG_ON(cpu_online(scpu
));
1655 tick_cancel_sched_timer(scpu
);
1657 local_irq_disable();
1658 old_base
= &per_cpu(hrtimer_bases
, scpu
);
1659 new_base
= this_cpu_ptr(&hrtimer_bases
);
1661 * The caller is globally serialized and nobody else
1662 * takes two locks at once, deadlock is not possible.
1664 raw_spin_lock(&new_base
->lock
);
1665 raw_spin_lock_nested(&old_base
->lock
, SINGLE_DEPTH_NESTING
);
1667 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1668 migrate_hrtimer_list(&old_base
->clock_base
[i
],
1669 &new_base
->clock_base
[i
]);
1672 raw_spin_unlock(&old_base
->lock
);
1673 raw_spin_unlock(&new_base
->lock
);
1675 /* Check, if we got expired work to do */
1676 __hrtimer_peek_ahead_timers();
1680 #endif /* CONFIG_HOTPLUG_CPU */
1682 static int hrtimer_cpu_notify(struct notifier_block
*self
,
1683 unsigned long action
, void *hcpu
)
1685 int scpu
= (long)hcpu
;
1689 case CPU_UP_PREPARE
:
1690 case CPU_UP_PREPARE_FROZEN
:
1691 init_hrtimers_cpu(scpu
);
1694 #ifdef CONFIG_HOTPLUG_CPU
1696 case CPU_DEAD_FROZEN
:
1697 migrate_hrtimers(scpu
);
1708 static struct notifier_block hrtimers_nb
= {
1709 .notifier_call
= hrtimer_cpu_notify
,
1712 void __init
hrtimers_init(void)
1714 hrtimer_cpu_notify(&hrtimers_nb
, (unsigned long)CPU_UP_PREPARE
,
1715 (void *)(long)smp_processor_id());
1716 register_cpu_notifier(&hrtimers_nb
);
1717 #ifdef CONFIG_HIGH_RES_TIMERS
1718 open_softirq(HRTIMER_SOFTIRQ
, run_hrtimer_softirq
);
1723 * schedule_hrtimeout_range_clock - sleep until timeout
1724 * @expires: timeout value (ktime_t)
1725 * @delta: slack in expires timeout (ktime_t)
1726 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1727 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1730 schedule_hrtimeout_range_clock(ktime_t
*expires
, unsigned long delta
,
1731 const enum hrtimer_mode mode
, int clock
)
1733 struct hrtimer_sleeper t
;
1736 * Optimize when a zero timeout value is given. It does not
1737 * matter whether this is an absolute or a relative time.
1739 if (expires
&& !expires
->tv64
) {
1740 __set_current_state(TASK_RUNNING
);
1745 * A NULL parameter means "infinite"
1752 hrtimer_init_on_stack(&t
.timer
, clock
, mode
);
1753 hrtimer_set_expires_range_ns(&t
.timer
, *expires
, delta
);
1755 hrtimer_init_sleeper(&t
, current
);
1757 hrtimer_start_expires(&t
.timer
, mode
);
1758 if (!hrtimer_active(&t
.timer
))
1764 hrtimer_cancel(&t
.timer
);
1765 destroy_hrtimer_on_stack(&t
.timer
);
1767 __set_current_state(TASK_RUNNING
);
1769 return !t
.task
? 0 : -EINTR
;
1773 * schedule_hrtimeout_range - sleep until timeout
1774 * @expires: timeout value (ktime_t)
1775 * @delta: slack in expires timeout (ktime_t)
1776 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1778 * Make the current task sleep until the given expiry time has
1779 * elapsed. The routine will return immediately unless
1780 * the current task state has been set (see set_current_state()).
1782 * The @delta argument gives the kernel the freedom to schedule the
1783 * actual wakeup to a time that is both power and performance friendly.
1784 * The kernel give the normal best effort behavior for "@expires+@delta",
1785 * but may decide to fire the timer earlier, but no earlier than @expires.
1787 * You can set the task state as follows -
1789 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1790 * pass before the routine returns.
1792 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1793 * delivered to the current task.
1795 * The current task state is guaranteed to be TASK_RUNNING when this
1798 * Returns 0 when the timer has expired otherwise -EINTR
1800 int __sched
schedule_hrtimeout_range(ktime_t
*expires
, unsigned long delta
,
1801 const enum hrtimer_mode mode
)
1803 return schedule_hrtimeout_range_clock(expires
, delta
, mode
,
1806 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range
);
1809 * schedule_hrtimeout - sleep until timeout
1810 * @expires: timeout value (ktime_t)
1811 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1813 * Make the current task sleep until the given expiry time has
1814 * elapsed. The routine will return immediately unless
1815 * the current task state has been set (see set_current_state()).
1817 * You can set the task state as follows -
1819 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1820 * pass before the routine returns.
1822 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1823 * delivered to the current task.
1825 * The current task state is guaranteed to be TASK_RUNNING when this
1828 * Returns 0 when the timer has expired otherwise -EINTR
1830 int __sched
schedule_hrtimeout(ktime_t
*expires
,
1831 const enum hrtimer_mode mode
)
1833 return schedule_hrtimeout_range(expires
, 0, mode
);
1835 EXPORT_SYMBOL_GPL(schedule_hrtimeout
);